语音端点检测算法

/****************************************************************************** * INCLUDE FILES ******************************************************************************/ #include <stdlib.h> #include <stdio.h> #include <math.h> #include <float.h> #include "amr_plus.h" /****************************************************************************** * PRIVATE PROGRAM CODE ******************************************************************************/ /****************************************************************************** * * Function : filter5 * Purpose : Fifth-order half-band lowpass/highpass filter pair with * decimation. * ******************************************************************************* */ static void filter5( float *in0, /* i/o : input values; output low-pass part */ float *in1, /* i/o : input values; output high-pass part */ float data[] /* i/o : updated filter memory */ ) { float temp0, temp1, temp2; temp0 = *in0 - COEFF5_1 * data[0]; temp1 = data[0] + COEFF5_1 * temp0; data[0] = temp0; temp0 = *in1 - COEFF5_2 * data[1]; temp2 = data[1] + COEFF5_2 * temp0; data[1] = temp0; *in0 = (temp1 + temp2)/2.0f; *in1 = (temp1 - temp2)/2.0f; } /****************************************************************************** * * Function : filter3 * Purpose : Third-order half-band lowpass/highpass filter pair with * decimation. * ******************************************************************************* */ static void filter3( float *in0, /* i/o : input values; output low-pass part */ float *in1, /* i/o : input values; output high-pass part */ float *data /* i/o : updated filter memory */ ) { float temp1, temp2; temp1 = *in1 - COEFF3 * *data; temp2 = *data + COEFF3 * temp1; *data = temp1; *in1 = (*in0 - temp2)/2.0f; *in0 = (*in0 + temp2)/2.0f; } /****************************************************************************** * * Function : level_calculation * Purpose : Calculate signal level in a sub-band. Level is calculated * by summing absolute values of the input data. * * Because speech coder has a lookahead, signal level calculated * over the lookahead (data[count1 - count2]) is stored (*sub_level) * and added to the level of the next frame. Additionally, group * delay and decimation of the filter bank is taken into the count * for the values of the counters (count1, count2). * ******************************************************************************* */ static float level_calculation( /* return: signal level */ float data[], /* i : signal buffer */ float *sub_level, /* i : level calculated at the end of the previous frame*/ /* o : level of signal calculated from the last */ /* (count2 - count1) samples */ Word16 count1, /* i : number of samples to be counted */ Word16 count2, /* i : number of samples to be counted */ Word16 ind_m, /* i : step size for the index of the data buffer */ Word16 ind_a, /* i : starting index of the data buffer */ float scale /* i : scaling for the level calculation */ ) { double l_temp1, l_temp2; float level; Word16 i; l_temp1 = 0.0; for (i = count1; i < count2; i++) { l_temp1 += fabs(data[ind_m*i+ind_a]); } l_temp1 *= 2.0f; l_temp2 = l_temp1 + *sub_level/scale; *sub_level = (float)(l_temp1*scale); for (i = 0; i < count1; i++) { l_temp2 += 2.0f*fabs(data[ind_m*i+ind_a]); } level = (float)(l_temp2*scale); return level; } /****************************************************************************** * * Function : filter_bank * Purpose : Divide input signal into bands and calculate level of * the signal in each band * ******************************************************************************* */ static void filter_bank( VadVars *st, /* i/o : State struct */ float in[], /* i : input frame */ float level[] /* 0 : signal levels at each band */ ) { Word16 i; float tmp_buf[FRAME_LEN]; /* shift input 1 bit down for safe scaling */ for (i = 0; i < FRAME_LEN; i++) { tmp_buf[i] = in[i]/2.0f; } /* run the filter bank */ for (i = 0;i < FRAME_LEN/2; i++) { filter5(&tmp_buf[2*i],&tmp_buf[2*i+1],st->a_data5[0]); } for (i = 0;i < FRAME_LEN/4; i++) { filter5(&tmp_buf[4*i],&tmp_buf[4*i+2],st->a_data5[1]); filter5(&tmp_buf[4*i+1],&tmp_buf[4*i+3],st->a_data5[2]); } for (i = 0; i < FRAME_LEN/8; i++) { filter5(&tmp_buf[8*i], &tmp_buf[8*i+4], st->a_data5[3]); filter5(&tmp_buf[8*i+2], &tmp_buf[8*i+6], st->a_data5[4]); filter3(&tmp_buf[8*i+3],&tmp_buf[8*i+7],&st->a_data3[0]); } for (i = 0; i < FRAME_LEN/16; i++) { filter3(&tmp_buf[16*i+0], &tmp_buf[16*i+8], &st->a_data3[1]); filter3(&tmp_buf[16*i+4], &tmp_buf[16*i+12], &st->a_data3[2]); filter3(&tmp_buf[16*i+6], &tmp_buf[16*i+14], &st->a_data3[3]); } for (i = 0; i < FRAME_LEN/32; i++) { filter3(&tmp_buf[32*i+0], &tmp_buf[32*i+16], &st->a_data3[4]); filter3(&tmp_buf[32*i+8], &tmp_buf[32*i+24], &st->a_data3[5]); } /* calculate levels in each frequency band */ /* 4800 - 6400 Hz*/ level[11] = level_calculation(tmp_buf, &st->sub_level[11], FRAME_LEN/4-48, FRAME_LEN/4, 4, 1, 0.25); /* 4000 - 4800 Hz*/ level[10] = level_calculation(tmp_buf, &st->sub_level[10], FRAME_LEN/8-24, FRAME_LEN/8, 8, 7, 0.5); /* 3200 - 4000 Hz*/ level[9] = level_calculation(tmp_buf, &st->sub_level[9], FRAME_LEN/8-24, FRAME_LEN/8, 8, 3, 0.5); /* 2400 - 3200 Hz*/ level[8] = level_calculation(tmp_buf, &st->sub_level[8], FRAME_LEN/8-24, FRAME_LEN/8, 8, 2, 0.5); /* 2000 - 2400 Hz*/ level[7] = level_calculation(tmp_buf, &st->sub_level[7], FRAME_LEN/16-12, FRAME_LEN/16, 16, 14, 1.0); /* 1600 - 2000 Hz*/ level[6] = level_calculation(tmp_buf, &st->sub_level[6], FRAME_LEN/16-12, FRAME_LEN/16, 16, 6, 1.0); /* 1200 - 1600 Hz*/ level[5] = level_calculation(tmp_buf, &st->sub_level[5], FRAME_LEN/16-12, FRAME_LEN/16, 16, 4, 1.0); /* 800 - 1200 Hz*/ level[4] = level_calculation(tmp_buf, &st->sub_level[4], FRAME_LEN/16-12, FRAME_LEN/16, 16, 12, 1.0); /* 600 - 800 Hz*/ level[3] = level_calculation(tmp_buf, &st->sub_level[3], FRAME_LEN/32-6, FRAME_LEN/32, 32, 8, 2.0); /* 400 - 600 Hz*/ level[2] = level_calculation(tmp_buf, &st->sub_level[2], FRAME_LEN/32-6, FRAME_LEN/32, 32, 24, 2.0); /* 200 - 400 Hz*/ level[1] = level_calculation(tmp_buf, &st->sub_level[1], FRAME_LEN/32-6, FRAME_LEN/32, 32, 16, 2.0); /* 0 - 200 Hz*/ level[0] = level_calculation(tmp_buf, &st->sub_level[0], FRAME_LEN/32-6, FRAME_LEN/32, 32, 0, 2.0); } /****************************************************************************** * * Function : update_cntrl * Purpose : Control update of the background noise estimate. * ******************************************************************************* */ static void update_cntrl( VadVars *st, /* i/o : State structure */ float level[] /* i : sub-band levels of the input frame */ ) { Word16 i; float stat_rat; float num, denom; float alpha; /* if fullband pitch or tone hav